Managing VM Resources...

[BentleyMiller86]

Hi all,

 

I have a few sites that have VM's as their DC and MIS Servers sat on a nice beefy Host, Host's by rule are:

 

2 x 16 Core Xeon Processors

64GB RAM

4 NIC in 1 Team

 

A few of them offer really poor performance when in the GUI but data transfer rate for network shares, and logon / folder redirection are all running nicely, particularly poor are load times when I restart the VM's , one of my DC's often takes up to 20 Minutes to load and iSCSI Backup data transfer rates, incremental backups are OK but the 2 week full backup often takes more than 48 Hours to complete a 1.2TB Backup, I i usually set resources as follows:

 

DC - 16 Cores and 32GB RAM

MIS - 4 Cores and 16GB RAM

 

Leaving 12 Cores and 12GB RAM for the Host.

 

Am I over doing it with the resources? And would it have a bearing on my perfornamce issues above?

 

Thanks

[ChrisC]

What does performance monitoring tell you in terms of load on the resources (for both VMs and physical)?

 

It almost sounds like the performance bottleneck could be the storage or network, are there any clues in event viewer to performance issues?

 

How beefy is the switching gear the host is connected to/the hops to backup?

[XiJ]

Mechanical drives? We try to use ssd / nvme for the drives the VM’s run on. Going to back to the mechanical hosts is horrible after that.

 

32gb seems a lot for a DC to me too - but how’s it depends on load and users etc.

[Jcx500]

My Dc's run fine on 4gb ram and 4 cores,Mis has 8gb and 4 cores,Lower level vm's to have single or two cores and 4gb of ram .Host has 128gb ddr4 and x2 xeon 4114 and all local storage

[BentleyMiller86]

What does performance monitoring tell you in terms of load on the resources (for both VMs and physical)?

It almost sounds like the performance bottleneck could be the storage or network, are there any clues in event viewer to performance issues?

How beefy is the switching gear the host is connected to/the hops to backup?

 

Hi Chris,

 

There's nothing in EV to indicate any issues, in terms of Load DC uses 25% of RAM and 0-10% of CPU, MIS uses 50-80% of RAM and 10-20% of CPU so there more then comfortable. All switching is 10/100/1000 HP un-managed

 

Mechanical drives? We try to use ssd / nvme for the drives the VM’s run on. Going to back to the mechanical hosts is horrible after that.

 

32gb seems a lot for a DC to me too - but how’s it depends on load and users etc.

 

Hi XiJ,

 

Yes they are mechanical, unfortunately SSD's are one of the first things to go when savings are needed to be made! however its a standard build I have the same setup in other sites that do not replicate the issues I have at others.

 

My Dc's run fine on 4gb ram and 4 cores,Mis has 8gb and 4 cores,Lower level vm's to have single or two cores and 4gb of ram .Host has 128gb ddr4 and x2 xeon 4114 and all local storage

 

Hi Jcx500,

 

The majority of my sites already use more resources than this, It may be possible when SIMS moves to the cloud, but not while its local.

Edited July 10, 2019 by BentleyMiller86

[Norphy]

16 cores and 32GB of RAM for a DC is waaaaaay overkill. I assign my DCs two cores and 4GB RAM. Even the ones for our busiest sites don't exceed 2GB of RAM being used and CPU usage is practically non-existent.

 

It's also generally a waste of time to assign more than four cores to a VM. I usually don't assign more than two unless the situation calls for it.

[ChrisC]

Hi Chris,

 

There's nothing in EV to indicate any issues, in terms of Load DC uses 25% of RAM and 0-10% of CPU, MIS uses 50-80% of RAM and 10-20% of CPU so there more then comfortable. All switching is 10/100/1000 HP un-managed

 

 

 

Hi XiJ,

 

Yes they are mechanical, unfortunately SSD's are one of the first things to go when savings are needed to be made! however its a standard build I have the same setup in other sites that do not replicate the issues I have at others.

 

 

Hi Jcx500,

 

The majority of my sites already use more resources than this, It may be possible when SIMS moves to the cloud, but not while its local.

What's the general age of the hosts?

 

*making some assumptions*

Are there any firmware updates available for the raid controller or hosts in general that could have some performance improvements possibly?

 

Are all hops between hosts and backup infrastructure connected at 1000?

 

I totally agree the specs allocated seem overkill and echo the suggestions that VMs generally need 4 cores max (especially as they're only running <10% utilisation, that also frees up additional resources should you need additional VMs in the future).

[BentleyMiller86]

What's the general age of the hosts?

 

*making some assumptions*

Are there any firmware updates available for the raid controller or hosts in general that could have some performance improvements possibly?

 

Are all hops between hosts and backup infrastructure connected at 1000?

 

I totally agree the specs allocated seem overkill and echo the suggestions that VMs generally need 4 cores max (especially as they're only running <10% utilisation, that also frees up additional resources should you need additional VMs in the future).

Hi Chris,

 

These hosts are new, less than a year.

 

I will check for Firmware updates again during the summer holidays but they were all up to date the last time I checked.

 

All Hops are at 1000, I made it my mission to have all 10/100 switching and connections sorted before the summer holidays, the only 100 connections I have now are to printers which still don't have 10/100/1000 NIC's in them, not much I can do about that really.

 

I may lower the resources to see if it makes any difference, but the likelihood of adding more hosts is very slim, Sophos, SIMS are now / going to be cloud based so its more likely I will be turning VM's off.

 

Thanks

Edited July 10, 2019 by BentleyMiller86